Gray Wolf proposes decentralized town-level data centers
Gray Wolf founder Pete Sacco argues that the next major AI market is local inference, requiring smaller, decentralized data centers rather than massive centralized training clusters. He contends that the current neocloud model is architecturally flawed due to mismatches between long-term leases and short-term GPU lifespans.
Pete Sacco, founder of Gray Wolf, argues that the next major AI market is inference, which does not require 100% GPU training stacks. He projects a shift toward smaller, decentralized data centers in every town, ranging from 300 kilowatt containers to 20 megawatt facilities, to handle local inference for AI devices and robots. Sacco stated, "I project we are going to have a data center in every town."
Sacco criticizes the neocloud model, exemplified by CoreWeave and Lambda, which signed 20-year leases on co-location space and financed GPUs with a 2-3 year lifespan. He argues this creates architectural risk if GPU dominance wanes. Instead, he advocates for a 'franchise' model of decentralization where local sites have sovereignty over data location, while governance, cybersecurity, and identification remain centralized.
Sacco notes that current enterprise trends involve building small language models on protected private information within IT data centers to avoid exposing data to public foundation models. He believes that decentralized, community-owned inference infrastructure is the necessary evolution for data sovereignty and efficiency, addressing the limitations of centralized models.